The Reliance Negotiation Framework: A Dynamic Process Model of Student LLM Engagement in Academic Writing

Shahin Hossain

arxiv: 2604.16772 · v1 · submitted 2026-04-18 · 💻 cs.CY · cs.AI· cs.HC

The Reliance Negotiation Framework: A Dynamic Process Model of Student LLM Engagement in Academic Writing

Shahin Hossain This is my paper

Pith reviewed 2026-05-10 07:31 UTC · model grok-4.3

classification 💻 cs.CY cs.AIcs.HC

keywords reliance negotiationLLM engagementacademic writingprocess modelAI ethicsmixed-methodsstudent decision-makingethical non-use

0 comments

The pith

Student engagement with LLMs in academic writing is a dynamic negotiation among four inputs that updates with each decision.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper claims that how undergraduates use large language models for writing is not a fixed characteristic but a process that unfolds differently for each assignment. Perceived benefits, risks, ethical commitments, and situational demands are weighed together, and the result of one negotiation alters how the next one proceeds. This approach accounts for why some students avoid LLMs altogether on ethical grounds and why more experience does not always improve outcomes. Data from 382 students at one institution support identifying these mechanisms and separating a non-negotiating group. The resulting framework offers predictions for how teaching and policies can respond to this variability.

Core claim

The Reliance Negotiation Framework reconceptualizes student LLM reliance in academic writing as an ongoing negotiation among four concurrent inputs: perceived benefits, perceived risks, ethical commitments, and situational demands. Outputs from each negotiation recursively modify the inputs for future decisions. The framework includes a Two-Model Architecture to handle the 13 percent of students whose ethical commitments rule out any use of LLMs without negotiation.

What carries the argument

The Reliance Negotiation Framework (RNF) as a dynamic process model of four inputs producing recursive outputs.

If this is right

Pedagogy in AI literacy should emphasize awareness of the four inputs and their interactions during writing.
Academic integrity policies should consider the situational and recursive aspects of reliance decisions.
Equity efforts can address how different students experience the negotiation process differently.
The model suggests that the developmental paradox of habituation can be mitigated by targeting feedback loops.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

This negotiation view may extend to student use of other generative AI tools in different subjects.
Future research could track the same students over multiple semesters to observe the recursive changes in real time.
Institutional policies might incorporate the framework to design support that helps shift negotiations toward lower risk or higher ethical alignment.

Load-bearing premise

The negotiation structure with its four inputs and recursive feedback, derived from one sample of undergraduates, applies as the general mechanism for LLM reliance in academic writing.

What would settle it

Finding that a static model based on individual differences like AI literacy or ethics scores predicts reliance better than the dynamic four-input negotiation across tasks would falsify the framework.

Figures

Figures reproduced from arXiv: 2604.16772 by Shahin Hossain.

**Figure 1.** Figure 1: The Reliance Negotiation Framework: four concurrent inputs (left) feed the negotiation process (center), producing three output dimensions (right). The amber dashed arc represents the recursive feedback loop through which each output modifies input weights governing subsequent negotiation cycles. The dashed horizontal boundary separates negotiation-mode students (87.0%, n = 314/361) from abstention-mode s… view at source ↗

read the original abstract

Student engagement with large language models (LLMs) in academic writing is not a stable trait, an adoption decision, or a competency level; it is a continuously negotiated process that existing frameworks cannot adequately theorize. Typological models provide categories without mechanisms; technology acceptance models explain adoption but not post-adoption quality; AI literacy frameworks treat competency as a static predictor rather than a live input. None accounts for within-student variability across tasks, the developmental paradox whereby experience produces habituation rather than sophistication, or principled non-use as a form of ethical reasoning. This article introduces the Reliance Negotiation Framework (RNF), developed from a sequential explanatory mixed-methods study of 382 undergraduates at a public minority-serving institution in the United States (survey, N = 382; 14 semi-structured interviews; three qualitative survey strands; 1,435 coded instances). The RNF reconceptualizes LLM reliance as an ongoing negotiation among four concurrent inputs (perceived benefits, perceived risks, ethical commitments, and situational demands) with outputs that recursively modify subsequent decisions. A Two-Model Architecture accommodates the 13.0% of participants whose categorical ethical commitments foreclose negotiation entirely. The framework generates four falsifiable predictions with implications for AI literacy pedagogy, academic integrity policy, and equity-centered practice at minority-serving institutions.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper introduces a recursive four-input negotiation model for student LLM use in writing, drawn from one MSI's mixed-methods data, but its claims to generality rest on untested scope.

read the letter

The main takeaway is that this work treats LLM reliance in academic writing as an ongoing negotiation among perceived benefits, risks, ethical commitments, and situational demands, with outputs feeding back to shape later choices. It also sets aside a separate path for the 13% of students whose ethical stance rules out any negotiation. That dynamic framing and the two-model split are positioned against typological buckets, static acceptance models, and fixed literacy views, which is the clearest point of departure from prior work. The mixed-methods base—382 surveys, 14 interviews, and 1,435 coded instances at a single public minority-serving institution—gives the framework some empirical footing and usefully centers equity questions at MSIs. The four falsifiable predictions are a reasonable next step if they can be tested independently. The main limitations are scope and verification. Everything comes from one institutional context, so the specific inputs and recursive structure may reflect local student demographics or culture rather than a general mechanism. The predictions appear derived from the same coded data, which raises the usual circularity concern unless the full paper shows separate validation steps. The abstract also omits sampling details, coding reliability, and exclusion criteria, leaving soundness hard to judge. This is worth a serious referee for people working on AI literacy, academic integrity policy, or higher-ed equity. Readers who need a mechanism account rather than categories could get value, but only after methods and generalizability are checked. Send it to peer review.

Referee Report

2 major / 1 minor

Summary. The paper introduces the Reliance Negotiation Framework (RNF) as a dynamic process model of student LLM engagement in academic writing. Drawing on a sequential explanatory mixed-methods study of 382 undergraduates at one public minority-serving institution (survey N=382, 14 semi-structured interviews, three qualitative survey strands, and 1,435 coded instances), the RNF models reliance as an ongoing negotiation among four concurrent inputs—perceived benefits, perceived risks, ethical commitments, and situational demands—whose outputs recursively modify subsequent decisions. It incorporates a Two-Model Architecture to handle the 13% of participants whose categorical ethical commitments foreclose negotiation, and derives four falsifiable predictions with implications for AI literacy pedagogy, academic integrity policy, and equity-centered practice.

Significance. If the central claims hold, the RNF would advance the field by supplying a mechanistic account of within-student variability, habituation effects, and principled non-use that existing typological, technology-acceptance, and static AI-literacy models lack. The mixed-methods design with a sizable sample and extensive coding, together with the explicit generation of falsifiable predictions, constitutes a concrete strength that could support empirical follow-up. The framework’s focus on minority-serving institutions also offers targeted implications for equity in AI-mediated writing instruction.

major comments (2)

[Abstract] Abstract: the description of the sequential explanatory mixed-methods study reports N=382, 14 interviews, and 1,435 coded instances but supplies no information on sampling strategy, exclusion criteria, inter-coder reliability, or the analytic steps by which the four inputs and recursive structure were induced from the data; because the framework is derived directly from these observations, the absence of these details leaves the validity of the central model unverified.
[Abstract and Discussion] Abstract and Discussion: the RNF and its Two-Model Architecture are presented as a general process model of LLM reliance, yet both were developed from a single-institution sample at one public minority-serving institution with no multi-site replication, cross-institutional comparison, or sampling that varies institutional type or regional context; this directly undermines the claim that the four-input negotiation constitutes a general mechanism rather than a site-specific pattern.

minor comments (1)

[Abstract] The manuscript introduces the 'Two-Model Architecture' in the abstract without a prior definition or diagram clarifying its components and how it interfaces with the main negotiation process.

Simulated Author's Rebuttal

2 responses · 0 unresolved

Thank you for the opportunity to respond to the referee's report. We address each major comment below with clarifications drawn from the full manuscript and indicate the revisions we will make.

read point-by-point responses

Referee: [Abstract] Abstract: the description of the sequential explanatory mixed-methods study reports N=382, 14 interviews, and 1,435 coded instances but supplies no information on sampling strategy, exclusion criteria, inter-coder reliability, or the analytic steps by which the four inputs and recursive structure were induced from the data; because the framework is derived directly from these observations, the absence of these details leaves the validity of the central model unverified.

Authors: We agree that the abstract should supply these details to support verification of the model's derivation. The full manuscript's Methods section provides them: sampling via convenience recruitment of undergraduates at the public minority-serving institution; exclusion of incomplete surveys and failed attention checks (yielding final N=382); inter-coder reliability via Cohen's kappa of 0.82 on a 20% double-coded subsample; and analytic steps consisting of open coding of interviews and qualitative survey strands followed by constant comparison to induce the four inputs and recursive structure. We will revise the abstract to include a brief summary of these elements. revision: yes
Referee: [Abstract and Discussion] Abstract and Discussion: the RNF and its Two-Model Architecture are presented as a general process model of LLM reliance, yet both were developed from a single-institution sample at one public minority-serving institution with no multi-site replication, cross-institutional comparison, or sampling that varies institutional type or regional context; this directly undermines the claim that the four-input negotiation constitutes a general mechanism rather than a site-specific pattern.

Authors: We acknowledge the single-institution sample as a genuine limitation that precludes treating the RNF as a fully general mechanism at this stage. The manuscript frames the framework as emerging from this specific context, with implications targeted at minority-serving institutions, and generates four falsifiable predictions precisely to support future multi-site testing. The Discussion already notes the need for broader validation. We will make a partial revision to both the abstract and Discussion to state this limitation more explicitly and to position the model as a context-derived starting point rather than an established general process. revision: partial

Circularity Check

0 steps flagged

No circularity; inductive framework development with forward-looking predictions

full rationale

The paper constructs the Reliance Negotiation Framework directly from its sequential explanatory mixed-methods study (N=382 survey, 14 interviews, 1,435 coded instances) at a single institution, identifying the four inputs and recursive structure empirically. It then states that this framework 'generates four falsifiable predictions.' No equations, fitted parameters renamed as predictions, or self-citation chains are present in the provided text. The predictions are presented as implications for future work rather than post-hoc re-statements tested on the same data. This follows standard qualitative theory-building practice and remains externally falsifiable, with no reduction of outputs to inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the assumption that the mixed-methods study validly extracts the four negotiation inputs and recursive structure from the 1,435 coded instances, plus the domain assumption that reliance is best modeled as negotiation rather than static traits.

axioms (1)

domain assumption A sequential explanatory mixed-methods design with survey and interview data can reliably identify the concurrent inputs and recursive outputs of LLM reliance.
The framework is constructed from the described study without independent validation of the coding or interpretation steps.

pith-pipeline@v0.9.0 · 5530 in / 1436 out tokens · 89719 ms · 2026-05-10T07:31:18.794257+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

20 extracted references · 20 canonical work pages

[1]

K., & Kendeou, P

Allen, L. K., & Kendeou, P. (2024). ED-AI Lit: An interdisciplinary framework for AI literacy in education. Policy Insights from the Behavioral and Brain Sciences, 11(1), 3--10. https://doi.org/10.1177/23727322231220339

work page doi:10.1177/23727322231220339 2024
[2]

Argyris, C., & Sch\" o n, D. A. (1978). Organizational learning: A theory of action perspective. Addison-Wesley

work page 1978
[3]

Atkinson, J. W. (1957). Motivational determinants of risk-taking behavior. Psychological Review, 64(6), 359--372. https://doi.org/10.1037/h0043445

work page doi:10.1037/h0043445 1957
[4]

Baidoo-Anu, D., & Ansah, L. O. (2023). Education in the era of generative artificial intelligence (AI): Understanding the potential benefits of ChatGPT in promoting teaching and learning. Journal of AI, 7(1), 52--62. https://doi.org/10.61969/jai.1337

work page doi:10.61969/jai.1337 2023
[5]

Barwise, J., & Perry, J. (1983). Situations and attitudes. MIT Press

work page 1983
[6]

Biggs, J. B. (1993). What do inventories of students' learning processes really measure? A theoretical review and clarification. British Journal of Educational Psychology, 63(1), 3--19. https://doi.org/10.1111/j.2044-8279.1993.tb01038.x

work page doi:10.1111/j.2044-8279.1993.tb01038.x 1993
[7]

Eaton, S. E. (2023). Artificial intelligence: Implications for academic integrity. Canadian Perspectives on Academic Integrity, 6(1), 1--17. https://doi.org/10.11575/cpai.v6i1.77015

work page doi:10.11575/cpai.v6i1.77015 2023
[8]

S., Wigfield, A., Harold, R., & Blumenfeld, P

Eccles, J. S., Wigfield, A., Harold, R., & Blumenfeld, P. B. (1993). Age and gender differences in children's self- and task perceptions during elementary school. Child Development, 64(3), 830--847. https://doi.org/10.2307/1131221

work page doi:10.2307/1131221 1993
[9]

S., Wigfield, A., Flanagan, C., Miller, C., Reuman, D., & Yee, D

Eccles, J. S., Wigfield, A., Flanagan, C., Miller, C., Reuman, D., & Yee, D. (1983). Expectancies, values, and academic behaviors. In J. T. Spence (Ed.), Achievement and achievement motivation (pp. 75--146). W. H. Freeman

work page 1983
[10]

Gibson, J. J. (1986). An ecological approach to visual perception. Lawrence Erlbaum. (Original work published 1979)

work page 1986
[11]

Greeno, J. G. (1998). The situativity of knowing, learning, and research. American Psychologist, 53(1), 5--26. https://doi.org/10.1037/0003-066X.53.1.5

work page doi:10.1037/0003-066x.53.1.5 1998
[12]

Hossain, S. (2026). Measuring college students' reliance on large language models in academic writing: A sequential explanatory mixed-methods study at a minority-serving institution [Doctoral dissertation, University of Maryland, Baltimore County]. UMBC Institutional Repository

work page 2026
[13]

Kohlberg, L. (1981). The philosophy of moral development: Moral stages and the idea of justice. Harper & Row

work page 1981
[14]

Lave, J., & Wenger, E. (1991). Situated learning: Legitimate peripheral participation. Cambridge University Press

work page 1991
[15]

Long, D., & Magerko, B. (2020). What is AI literacy? Competencies and design considerations. In Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems (pp. 1--16). Association for Computing Machinery. https://doi.org/10.1145/3313831.3376727

work page doi:10.1145/3313831.3376727 2020
[16]

Meehl, P. E. (1978). Theoretical risks and tabular asterisks: Sir Karl, Sir Ronald, and the slow progress of soft psychology. Journal of Consulting and Clinical Psychology, 46(4), 806--834. https://doi.org/10.1037/0022-006X.46.4.806

work page doi:10.1037/0022-006x.46.4.806 1978
[17]

Nozick, R. (1974). Anarchy, state, and utopia. Basic Books

work page 1974
[18]

Strzelecki, A. (2024). Students' acceptance of ChatGPT in higher education: An extended unified theory of acceptance and use of technology. Innovative Higher Education, 49(1), 223--245. https://doi.org/10.1007/s10755-023-09686-1

work page doi:10.1007/s10755-023-09686-1 2024
[19]

Wigfield, A., & Eccles, J. S. (2000). Expectancy--value theory of achievement motivation. Contemporary Educational Psychology, 25(1), 68--81. https://doi.org/10.1006/ceps.1999.1015

work page doi:10.1006/ceps.1999.1015 2000
[20]

S., Yoon, K

Wigfield, A., Eccles, J. S., Yoon, K. S., Harold, R. D., Arbreton, A. J. A., Freedman-Doan, C., & Blumenfeld, P. C. (1997). Change in children's competence beliefs and subjective task values across the elementary school years: A three-year study. Journal of Educational Psychology, 89(3), 451--469. https://doi.org/10.1037/0022-0663.89.3.451

work page doi:10.1037/0022-0663.89.3.451 1997

[1] [1]

K., & Kendeou, P

Allen, L. K., & Kendeou, P. (2024). ED-AI Lit: An interdisciplinary framework for AI literacy in education. Policy Insights from the Behavioral and Brain Sciences, 11(1), 3--10. https://doi.org/10.1177/23727322231220339

work page doi:10.1177/23727322231220339 2024

[2] [2]

Argyris, C., & Sch\" o n, D. A. (1978). Organizational learning: A theory of action perspective. Addison-Wesley

work page 1978

[3] [3]

Atkinson, J. W. (1957). Motivational determinants of risk-taking behavior. Psychological Review, 64(6), 359--372. https://doi.org/10.1037/h0043445

work page doi:10.1037/h0043445 1957

[4] [4]

Baidoo-Anu, D., & Ansah, L. O. (2023). Education in the era of generative artificial intelligence (AI): Understanding the potential benefits of ChatGPT in promoting teaching and learning. Journal of AI, 7(1), 52--62. https://doi.org/10.61969/jai.1337

work page doi:10.61969/jai.1337 2023

[5] [5]

Barwise, J., & Perry, J. (1983). Situations and attitudes. MIT Press

work page 1983

[6] [6]

Biggs, J. B. (1993). What do inventories of students' learning processes really measure? A theoretical review and clarification. British Journal of Educational Psychology, 63(1), 3--19. https://doi.org/10.1111/j.2044-8279.1993.tb01038.x

work page doi:10.1111/j.2044-8279.1993.tb01038.x 1993

[7] [7]

Eaton, S. E. (2023). Artificial intelligence: Implications for academic integrity. Canadian Perspectives on Academic Integrity, 6(1), 1--17. https://doi.org/10.11575/cpai.v6i1.77015

work page doi:10.11575/cpai.v6i1.77015 2023

[8] [8]

S., Wigfield, A., Harold, R., & Blumenfeld, P

Eccles, J. S., Wigfield, A., Harold, R., & Blumenfeld, P. B. (1993). Age and gender differences in children's self- and task perceptions during elementary school. Child Development, 64(3), 830--847. https://doi.org/10.2307/1131221

work page doi:10.2307/1131221 1993

[9] [9]

S., Wigfield, A., Flanagan, C., Miller, C., Reuman, D., & Yee, D

Eccles, J. S., Wigfield, A., Flanagan, C., Miller, C., Reuman, D., & Yee, D. (1983). Expectancies, values, and academic behaviors. In J. T. Spence (Ed.), Achievement and achievement motivation (pp. 75--146). W. H. Freeman

work page 1983

[10] [10]

Gibson, J. J. (1986). An ecological approach to visual perception. Lawrence Erlbaum. (Original work published 1979)

work page 1986

[11] [11]

Greeno, J. G. (1998). The situativity of knowing, learning, and research. American Psychologist, 53(1), 5--26. https://doi.org/10.1037/0003-066X.53.1.5

work page doi:10.1037/0003-066x.53.1.5 1998

[12] [12]

Hossain, S. (2026). Measuring college students' reliance on large language models in academic writing: A sequential explanatory mixed-methods study at a minority-serving institution [Doctoral dissertation, University of Maryland, Baltimore County]. UMBC Institutional Repository

work page 2026

[13] [13]

Kohlberg, L. (1981). The philosophy of moral development: Moral stages and the idea of justice. Harper & Row

work page 1981

[14] [14]

Lave, J., & Wenger, E. (1991). Situated learning: Legitimate peripheral participation. Cambridge University Press

work page 1991

[15] [15]

Long, D., & Magerko, B. (2020). What is AI literacy? Competencies and design considerations. In Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems (pp. 1--16). Association for Computing Machinery. https://doi.org/10.1145/3313831.3376727

work page doi:10.1145/3313831.3376727 2020

[16] [16]

Meehl, P. E. (1978). Theoretical risks and tabular asterisks: Sir Karl, Sir Ronald, and the slow progress of soft psychology. Journal of Consulting and Clinical Psychology, 46(4), 806--834. https://doi.org/10.1037/0022-006X.46.4.806

work page doi:10.1037/0022-006x.46.4.806 1978

[17] [17]

Nozick, R. (1974). Anarchy, state, and utopia. Basic Books

work page 1974

[18] [18]

Strzelecki, A. (2024). Students' acceptance of ChatGPT in higher education: An extended unified theory of acceptance and use of technology. Innovative Higher Education, 49(1), 223--245. https://doi.org/10.1007/s10755-023-09686-1

work page doi:10.1007/s10755-023-09686-1 2024

[19] [19]

Wigfield, A., & Eccles, J. S. (2000). Expectancy--value theory of achievement motivation. Contemporary Educational Psychology, 25(1), 68--81. https://doi.org/10.1006/ceps.1999.1015

work page doi:10.1006/ceps.1999.1015 2000

[20] [20]

S., Yoon, K

Wigfield, A., Eccles, J. S., Yoon, K. S., Harold, R. D., Arbreton, A. J. A., Freedman-Doan, C., & Blumenfeld, P. C. (1997). Change in children's competence beliefs and subjective task values across the elementary school years: A three-year study. Journal of Educational Psychology, 89(3), 451--469. https://doi.org/10.1037/0022-0663.89.3.451

work page doi:10.1037/0022-0663.89.3.451 1997